common/snprintf.c: preserve wide integer values

[user01010111]

Summary

• Preserve fallback integer formatting at LLONG width.
• Prevent %p and %ll* values from being truncated through long on LLP64 systems.
• Handle LLONG_MIN without signed overflow.
• Add focused TEST_SNPRINTF comparisons for pointers, wide %lld values and LLONG_MIN.

Fixes #1602

Validation

• Native Windows 11 x64, MSYS2 MinGW64 GCC 16.2.0, LLP64:
  • the upstream fallback fails under -std=c99 -Wall -Werror
  • the candidate compiles warning-clean
  • the old formatter fails exactly the three added %p, wide %lld and LLONG_MIN regression cases
  • the candidate passes all 191 TEST_SNPRINTF comparisons
• Native Windows Autotools integration:
  • snprintf and vsnprintf were forced unavailable through configure
  • make -C common V=1 snprintf.lo passes under -Werror
• Native GCC 16.2.1 and Clang 22.1.8: 191/191 comparisons passed.
• GCC UBSan: 191/191 comparisons passed.
• make stylecheck, the non-ASCII source check and git diff --check passed.

General C checklist

• Integer widths and value ranges are handled explicitly rather than assuming long matches pointer or long-long width.
• Direct printf and sprintf declarations remain isolated to the existing TEST_SNPRINTF harness; production logging behaviour is unchanged.
• Coding style and whitespace follow common/snprintf.c and docs/developers.txt precedent.
• No files were added, so no build or distribution lists need updating.

AI assistance

OpenAI Codex gpt-5.6-sol was used for repository analysis, implementation, review, drafting and native/local validation. The human contributor reviewed the complete diff and validation evidence and remains responsible for the change.

[github-actions]

A ZIP file with standard source tarball and another tarball with pre-built docs for commit 17548a3 is temporarily available: NUT-tarballs-PR-3598.zip.

[user01010111]

The CentOS 8/ppc64le OBS failure occurs during RPM database initialisation, before %prep or any source compilation:

BDB0091 DB_VERSION_MISMATCH: Database environment version mismatch

CentOS 9/ppc64le builds this commit successfully. Could the CentOS 8/ppc64le preinstall image (a0769ebbc600e42fa9742d61cacaedd7) be refreshed before rerunning the target?

[AppVeyorBot]

✅ Build nut 2.8.5.5161-master completed (commit ee81f6e695 by @)

• artifacts
• A 7-zip archive with built NUT for Windows debug binaries is temporarily available: NUT-for-Windows-x86_64-SNAPSHOT-2.8.5.5161-master.7z

[AppVeyorBot]

✅ Build nut 2.8.5.5161-master completed (commit ee81f6e695 by @)

[jimklimov]

OBS scenarios are on Open Build System, following their dependency tree etc. - as often as they regenerate them. I gather there is a regular inability to start the (emulated?) builders for this platform, which is mostly worked around by persistent restarting of the build in their Web-UI...

[jimklimov]

I think the signed number support got broken here, the rest seems OK, thanks.

[jimklimov]

This one change does not look right.

[user01010111]

Thanks, Jim. This is intentional unsigned arithmetic. After the negative LLONG value is converted to the corresponding unsigned type, C defines unary minus modulo 2^N, so this produces its magnitude without overflowing for LLONG_MIN. Using -value here would be undefined for LLONG_MIN. The added %lld comparison for LLONG_MIN exercises this path and passed against the system formatter.

Happy to add a short explanatory comment if that would make the intent clearer?

[jimklimov]

I think yes, to avoid same questions later on. "Bits magic, not maths magic" :)

[user01010111]

Thanks, Jim. Fair point: unsigned negation in shared formatting code is unusual enough to deserve scrutiny. I've added a short comment so future readers do not have to reconstruct the modulo arithmetic to see why it is intentional.

[jimklimov]

Still, I wonder how many platforms actually build this fallback implementations.

From experience with annoying compiler warnings, I'd actually expect this one with neither range-checks nor explicit casting to raise questions from static analysis.

Wouldn't it be clearer somewhat like this?

if (flags & DP_F_UNSIGNED) {
    uvalue = (unsigned LLONG)value;
} else {
    if (value < 0) {
        signvalue = '-';
        uvalue = -value;
    } else {
        uvalue = value;
        if (flags & DP_F_PLUS)  /* Do a sign (+/i) */
            signvalue = '+';
        else
            if (flags & DP_F_SPACE)
                signvalue = ' ';
    }
}